Provenance over volume: why coverage is not confirmation
A claim that appears in fifty places can still rest on a single source. Counting the places tells you about distribution, not about truth. The job is to find the origin and weigh that.
Eight outlets reprinting one wire story are one piece of evidence, not eight.
Coverage is a distribution signal, not a truth signal
When a figure spreads across the web, most of what you are seeing is syndication. A wire service publishes once, and outlets reprint it with light edits. The figure is now in many places, but it has been checked once, if at all.
Treating each appearance as independent confirmation is how a single unsupported number ends up looking like consensus. The volume is real; the agreement is not.
Cluster first, then weigh
Before any evidence counts toward a claim, retrieved passages are grouped by their underlying source. Near duplicate text is detected and merged, and items that trace to the same origin form one cluster.
The cluster contributes once, carrying the weight of its strongest member. The reprints are kept in the audit trail so a reviewer can see the spread, but they do not move the number on their own.
What this changes in practice
A reading stops inflating when a story goes viral. A primary filing outweighs a hundred summaries of it. And when the only support for a claim is a cloud of reprints with no traceable origin, the claim is flagged rather than trusted.
Takeaways
- Evidence is grouped by origin before it counts.
- A cluster contributes once, at the weight of its strongest member.
- Reprints stay in the trail for review, not in the score.